To measure PCP identity we use the MetaNeighbor method as described in our companion paper (Crow et. al. 2017). In brief, MetaNeighbor requires the input of a set of genes, an expression matrix and two sets of labels: one set for labeling each experiment, and one set for labeling the cell types of interest. We perform a stratified cross-validation which allows us to explicitly block technical sources of variation in single-cell analysis, in close parallel to our meta-analytic evaluation of single-cell data (Crow et. al. 2017). Here, each batch was treated as an “experiment”, and we aimed to measure the replicability of cell identity across batches. Cell-type labels are held back from one experiment at a time and then predicted based on the others, to determine which gene sets functionally characterize cells across technical variation. For each gene set being used to evaluate a given cell-type, the method generates a network based on the Spearman correlation between all cells across the genes within the set. The correlation is rank standardized to provide network weightings between each pair of cells, and
